Band gap prediction and interpretation method based on interpretable machine learning

Abstract

The invention discloses a band gap prediction and interpretation method based on interpretable machine learning, and the method comprises the steps: collecting double perovskite oxide data, and obtaining a first data set; preprocessing the first data set to obtain a second data set; selecting a feature subset of physical features related to the band gap from the second data set; setting parameters of a random forest algorithm, and constructing a prediction model; dividing the second data set into a training set and a test set; performing prediction training on the data of the training set, and detecting a prediction effect through the test set; and performing interpretable analysis on the second data set and the feature subset. According to the method, the defects of a traditional experimental and theoretical calculation method are overcome, time and resources are saved, and a model result can be quickly obtained by inputting related data into a model which is established based on interpretable machine learning and is used for predicting and interpreting the band gap of the double perovskite oxide.

Description

Technical Field

[0001] This application belongs to the technical field of computational materials, and particularly relates to a bandgap prediction and interpretation method based on interpretable machine learning. Background Art

[0002] Double perovskite oxide materials have received extensive attention due to their excellent optoelectronic properties. The chemical general formula of double perovskite oxides is A 2 B 2 O 6 . Among them, the A-site is usually a metal element with a relatively large ionic radius such as rare earth or alkaline earth, which coordinates with 12 oxygen atoms to form a closest cubic packing and plays a role in stabilizing the structure; the B-site is generally a transition metal element, which coordinates with 6 oxygen atoms and occupies the octahedral center in the cubic close packing. In solid oxide fuel cells, double perovskite oxides can be candidates for new cathode and anode materials to improve the output power and stability of the battery. Due to the specific electronic structure and optical properties of double perovskite oxides, they can be applied to optoelectronic devices such as photodetectors and photocatalysts.

[0003] The bandgap refers to the energy interval between the valence band and the conduction band in a solid material. In solid physics, the energy levels of electrons are not continuous but are distributed in different energy bands. There are some energy regions where electrons are prohibited from existing between these energy bands, which is the bandgap. The size and nature of the bandgap directly determine the electrical, optical and other physical properties of the solid. The bandgap of perovskite materials is an important physical property, which determines the application range of the materials in optoelectronic devices. A remarkable feature of perovskite materials is that their bandgaps have the characteristic of continuous tunability. By changing the chemical composition of the material, the adjustment from narrow bandgap to wide bandgap can be achieved.

[0004] With the continuous development of artificial intelligence technology, more and more scientific researchers have begun to introduce machine learning technology into the field of materials science. Facing a large amount of experimental data and theoretical calculation results, they use machine learning methods to predict various properties of materials. This not only greatly reduces the resources required for experiments and calculations, but also provides strong guidance for experimental research and practical applications. Summary of the Invention

[0005] This application provides a bandgap prediction and interpretation method based on interpretable machine learning to solve the above technical problems.

[0006] To solve the above technical problems, a technical solution adopted in this application is: a bandgap prediction and interpretation method based on interpretable machine learning, including:

[0007] Collect double perovskite oxide data to obtain a first data set; the double perovskite oxide data set includes material chemical formulas, bandgaps, and physical characteristics;

[0008] Preprocess the first dataset to obtain the second dataset;

[0009] Select a feature subset of physical features related to the bandgap from the second dataset;

[0010] Based on the random forest algorithm, set the parameters of the random forest algorithm and construct a prediction model;

[0011] Divide the second dataset into a training set and a test set;

[0012] Based on the prediction model, perform prediction training on the data in the training set and detect the prediction effect through the test set;

[0013] Conduct interpretable analysis on the second dataset and the feature subset.

[0014] Furthermore, the preprocessing method includes:

[0015] Delete the double perovskite oxidation data in the first dataset with missing feature variables greater than the threshold; among them, the double perovskite oxidation data with missing feature variables less than the threshold in the first dataset is replaced by taking the average value of the missing values

[0016] Furthermore, the method of dividing the second dataset into a training set and a test set includes:

[0017] By setting a random factor, divide the second dataset into a training set and a test set according to a ratio of 4:1 each time.

[0018] Furthermore, the prediction training method includes:

[0019] Perform 10 random splits on the data in the training set, calculate and print the average value of the R 2 score and RMSE score after each random split.

[0020] Furthermore, the method of conducting interpretable analysis on the second dataset and the feature subset includes:

[0021] Based on the second dataset and the feature subset, calculate the shap value of the feature subset, obtain the feature importance ranking, and use it as an explanation for the bandgap of the double perovskite oxide.

[0022] The beneficial effects of this application are:

[0023] 1. This application overcomes the deficiencies of traditional experimental and theoretical calculation methods, saves time and resources, and based on an interpretable machine learning-based model for predicting and explaining the bandgap of double perovskite oxides, the relevant data can be input into the model to quickly obtain the model results.

[0024] 2. The prediction and interpretation method of the present application does not require experimental operations during implementation and does not need to rely on any chemical reagents. Therefore, it will not cause chemical pollution, conforms to the development concept of environmental protection, is easy to operate, and is easy to implement, with the potential for wide promotion and application.

[0025] 3. The present application has the ability to predict and interpret the band gap of double perovskite oxides, which can help experimental researchers accurately screen samples that meet specific conditions for experimental verification, thereby saving a large amount of experimental time and resources, improving experimental efficiency, playing an important guiding role, and reducing blindness during the experiment. In addition, the method of the present invention is easy to operate and has low cost, with wide promotion and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a schematic flowchart of an embodiment of the band gap prediction and interpretation method based on interpretable machine learning of the present application;

[0027] Figure 2 is a prediction result diagram of the band gap of double perovskite oxide materials by the random forest algorithm of the present application;

[0028] Figure 3 is an importance ranking diagram of the band gap characteristics of double oxide perovskite calculated by the random forest algorithm of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] To make the objectives, technical solutions, and advantages of the present invention clearer, the following further elaborates on the present invention in detail with reference to specific embodiments.

[0030] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the specific embodiments disclosed in the following specification.

[0031] Refer to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of the band gap prediction and interpretation method based on interpretable machine learning of the present application. The method includes:

[0032] Step S1. Collect double perovskite oxide data to obtain a first data set; the double perovskite oxide data set includes material chemical formulas, band gaps, and physical characteristics.

[0033] Specifically, collect double perovskite oxide chemical formulas, band gap data, and characteristic information from publicly available databases as a machine learning sample data set. There are 1,680 pieces of such data in total.

[0034] Step S2. Preprocess the first data set to obtain a second data set.

[0035] Specifically, preprocess the first data set and delete the double perovskite oxidation data with missing feature variables greater than the threshold in the first data set; among them, for the double perovskite oxidation data with missing feature variables less than the threshold in the first data set, the missing values are replaced by taking the average value.

[0036] In this application, the material data with multiple missing feature variables is deleted row by row, and the data with a single missing feature variable is filled by taking the average value. Finally, 1305 data are obtained.

[0037] Step S3. Select a feature subset of physical features related to the band gap from the second data set.

[0038] Specifically, select 10 physical feature variables related to the band gap, as shown in Table 1;

[0039] Table 1 is a table of physical feature variables

[0040]

[0041] Step S4. Based on the random forest algorithm, set the parameters of the random forest algorithm and construct a prediction model.

[0042] Specifically, set the number of base learners, i.e., decision trees, in the random forest algorithm to 50, set the maximum depth of the tree to 10, the minimum number of samples required to split an internal node to 2, and the minimum number of samples required at a leaf node to 1. It should be noted that the random forest is an ensemble learning method that improves the accuracy and robustness of the model by constructing multiple decision trees and aggregating their prediction results. This application improves on the random forest algorithm so that it can not only predict the band gap of double perovskite oxides but also explain the band gap.

[0043] Step S5. Divide the second data set into a training set and a test set.

[0044] Specifically, by setting a random factor, the second data set is divided into a training set and a test set according to a ratio of 4:1 each time.

[0045] Step S6. Based on the prediction model, perform prediction training on the data in the training set and detect the prediction effect through the test set.

[0046] Specifically, use the random forest algorithm to construct a band gap prediction model for the second data set and the feature subset, randomly split the data in the training set 10 times, and calculate the R2 and RSME values of the model. The value of R 2 is 0.883, and the value of RSME is 0.535, as Figure 2 shown.

[0047] It should be noted that when calculating R2 When calculating the R and RSME values, to prevent the contingency of the results, this application calculates and prints the average values of the R scores and RMSE scores after these 10 segmentation iterations to evaluate the performance of the model. 2

[0048] Step S7. Conduct an interpretable analysis on the second data set and the feature subset.

[0049] Specifically, first calculate the shap values of the physical feature variables used in the random forest algorithm prediction model, and then obtain the feature ranking according to the shap values.

[0050] The coefficient of determination R of the model prediction of this application 2 is 0.883, and the root mean square error is RSME = 0.535. The method of this application predicts the band gap of the material through the data collected from the material database, reduces the costs of theoretical calculation and experiment, and improves the prediction speed of the band gap of double perovskite oxides.

[0051] Refer to Figure 3 , calculate the shap values of the physical feature variables used in the random forest algorithm model, and then calculate the feature importance ranking of the physical feature variables according to the obtained shap values. It is calculated that the average number of valence electrons on the d orbit of the constituent elements, the range of unfilled valence electrons in the constituent elements, and the highest melting point of the material constituent elements have the greatest influence on predicting the band gap of double perovskite oxides, and the results are as Figure 3 shown.

[0052] This application has established an efficient prediction and interpretation model through the experimental samples in the source database, which has the advantages of high accuracy, low cost, and environmental friendliness. And it can quickly guide the experimental synthesis of related materials, with the characteristics of fast calculation speed, high precision, and convenient electronic structure analysis.

[0053] The above are only the embodiments of this application, and do not limit the patent scope of this application accordingly. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of this application.

Claims

1. A bandgap prediction and interpretation method based on interpretable machine learning, characterized in that: include: collecting double perovskite oxide data to obtain a first data set; The double perovskite oxide data set includes material chemical formula, band gap, and physical characteristics; Preprocessing the first data set to obtain a second data set; selecting a subset of features of physical features related to the band gap from the second data set; Based on the random forest algorithm, set the parameters of the random forest algorithm and build a prediction model; Dividing the second data set into a training set and a test set; Based on the prediction model, prediction training is performed on the data of the training set, and the prediction effect is tested through the test set; An interpretable analysis is performed on the second data set and the feature subset.

2. The method according to claim 1, characterized in that The pretreatment method comprises: The double perovskite oxidation data in the first data set whose missing characteristic variables are greater than a threshold are deleted; wherein, the double perovskite oxidation data in the first data set whose missing characteristic variables are less than the threshold are replaced by averaging the missing values.

3. The method according to claim 1, characterized in that The method of dividing the second data set into a training set and a test set comprises: By setting a random factor, the second data set is divided into the training set and the test set at a ratio of 4:1 each time.

4. The method according to claim 3, characterized in that The prediction training method comprises: The data in the training set are randomly split 10 times, and the R value of the data after each random split is calculated and printed. 2 The average of the scores and RMSE scores.

5. The method according to claim 2, characterized in that: The method for performing interpretable analysis on the second data set and the feature subset comprises: Based on the second data set and the feature subset, the shap value of the feature subset is calculated to obtain the feature importance ranking as an explanation for the band gap of the double perovskite oxide.

Images